Informal talk by Michael Dorr, who is a student working with Erhardt Barth in Lubeck, Germany. The talk given on October 21, 2008 at the Redwood Center for Theoretical Neuroscience at UC Berkeley.
We predict where people will look in natural scenes using Machine Learning techniques on different video representations. One recent result is that sparser representations yield better predictability. Ultimately, we are incorporating these results into gaze-contingent displays that change the saliency of a scene in real time, in order to guide gaze and optimize visual communication.